MSnet: A Neural Network which Classifies Mass Spectra

Abstract

We have designed a feed-forward neural network to classify low-resolution mass spectra of unknown compounds according to the presence or absence of 100 organic substructures. The neural network, MSnet, was trained to compute a maximum-likelihood estimate of the probability that each substructure is present. We discuss some design considerations and statistical properties of neural network classifiers, and the effect of various training regimes on generalization behavior. The MSnet classifies mass spectra more reliably than other methods reported in the literature, and has other desirable properties.